About Plext

The problem

Your team uses AI coding agents every day. Each one starts every session from zero.

The bug your teammate fixed last Tuesday in a ninety-minute Cursor chat — gone. The reason you picked SST over Serverless Framework — buried in a Claude Code transcript that scrolled off the screen. The "don't ever do X with Y" lesson someone learned the hard way — re-learned by a different agent next week, same cost, same hour.

This is the new tax on AI-assisted development: every conversation produces value, and almost none of it survives the conversation. The agent that just helped you ship will have forgotten by morning. The next teammate's agent will start over.

What Plext is

A shared knowledge base that AI agents and humans both read from and write to, sitting outside any single agent or IDE. Notes live in plexes — named collections, optionally with schemas. Agents save what they learn. Other agents (and humans) find it via search before acting.

Four things make this practical:

Tool-agnostic. Same notes, every agent. Read and write from Claude Code, Cursor, Codex, your CI runner, or curl. Switch tools tomorrow and the knowledge stays put.
Two-way. The same data model serves human writes from the dashboard, agent writes via MCP / CLI, and programmatic writes via the API. One source of truth.
Schemas when you need them. A plex can require fields like severity, area, or status — with types validated on write. Or stay schemaless. The same plex can mix freeform tags with strict required fields.
Role-based by default. Every team member has a role; every action is gated by it. API keys inherit from the member who issued them and can narrow to specific plexes and to read or write independently. Demote a member, their tokens narrow on the very next request.

Why now

Coding agents — Claude Code, Cursor, Codex, Claude.ai — became real in 2025–2026. They need durable memory across sessions, across teammates, and across whichever tool you happen to be running this quarter.

When agents have shared memory, the bottleneck stops being what does the agent know and becomes what does the team know that the agent can find. That's a much easier problem.

No existing tool was built for this. Notion is for humans. Vector DBs are for retrieval, not curation. Wikis decay. Plext is the missing layer.

What we're building toward

A world where any agent on any team starts a task by reading what's already known, and ends by writing what was learned. Org knowledge that compounds by default — not because someone remembers to write it down, but because the tools the team uses every day make saving the path of least resistance.

What Plext is not

Not a chatbot. Plext doesn't generate answers. It stores them.
Not a vector store. We use embeddings for semantic search, but every note is plain text you can read, edit, and audit.
Not Notion. No nested pages, no databases, no inline calendars. A plex is a flat list of notes — that's the whole shape.
Not a wiki. Wikis assume humans curate. Plext assumes agents curate, with humans editing.

Where we are today

v1 API is stable. Bearer-token-only, fifteen endpoints, additive-only changes.
CLI ships skills for Claude Code, Codex CLI, and Cursor.
Hosted on plext.com. Sign up is free during the early-access period.